Orchestration for infrastructure that doesn't sit still.
Salad Container Engine (SCE) is the software layer that schedules, monitors, and reallocates container workloads across Community Cloud, Salad Dedicated, or a private fleet. Bring a Docker image, hardware requirements, and a replica count; SCE handles the distributed infrastructure underneath.
Billed per second, only while instances are running. Cold start and image download are free.
GPUs online right now
Loading…Machines already reserved on contract are not included in these counts.
Three inputs. Everything else is handled.
Bring a container
Any amd64 OCI image from Docker Hub, GHCR, AWS ECR, or a private registry. Credentials are used once to cache the image, then discarded.
Choose hardware and priority
Pick GPU classes, vCPU, and RAM. Choose High, Medium, Low, or Lowest priority to trade interruption tolerance for price.
Set replicas and go
SCE matches your requirements to the best available nodes using the trust rating, starts instances, and keeps the replica count filled as nodes come and go.
Built for distributed fleets.
SCE was designed to make large, varied fleets behave like dependable infrastructure, whether the nodes come from Community Cloud, reserved hardware, or your own organization.
Trust-rated scheduling
A proprietary trust rating indexes each node's performance and availability history. Workloads land on the most reliable hardware that meets their spec.
Automatic reallocation
When a node disconnects, SCE reschedules the instance to another node of the same class. You set the replica count; SCE keeps it filled.
Hypervisor isolation
Containers run inside a Linux VM partitioned from the host by Hyper-V. Isolation works both ways: your workload is protected from the host, and the host from your workload.
Container Gateway
Expose a port through a WireGuard-backed gateway with TLS, load balancing, and DDoS protection. One replica per group gets a stable, unique URL.
Job Queues
Push work to a queue and let SCE distribute it across replicas with webhook callbacks. Completed jobs are deleted after 24 hours.
Health probes and logs
Startup, liveness, and readiness probes; heartbeat monitoring; log streaming to Axiom, Datadog, New Relic, Splunk, or any HTTP endpoint.
Registries and secrets
Images and environment variables encrypted at rest and in transit. Private registry credentials never persist on nodes.
API, SDKs, and Kubernetes
A full REST API with Python and TypeScript SDKs, plus a Virtual Kubelet so existing K8s pipelines can schedule onto SaladCloud unchanged.
SSH and web terminal
Debug a running instance over SSH or a browser terminal with keys managed in the portal.
The same SCE runs everywhere Salad runs.
16,000+ daily active consumer GPUs
The lowest prices anywhere for stateless, interruption-tolerant inference and batch. RTX 3060 to RTX 5090.
Community Cloud →Reserved servers and VMs
Pin container groups to dedicated RTX PRO 6000 Blackwell hardware on contract for predictable capacity and larger VRAM.
Salad Dedicated →Your organization's idle GPUs
Deploy the SCE control plane inside your network and schedule onto the workstations and servers you already own.
Private Cloud →Production, not proof of concept.
"By switching to SaladCloud, we are serving inference on over 600 consumer GPUs to deliver 10 Million images per day and training more than 15,000 LoRAs per month. SaladCloud not only had the lowest GPU prices in the market but also offered us incredible scalability."
Justin Maier · Founder & CEO, CivitaiRead the case study →"On SaladCloud's consumer GPUs, we are running 3X more scale at half the cost of A100s on our local provider and almost 85% less cost than the two major hyperscalers we were using before. I'm not losing sleep over scaling issues anymore."
Jamsheed Kamardeen · CTO, BlendRead the case study →Before you deploy.
What do I pay for?
Running time only, per second. Allocation, image download, and cold start are not billed. GPU instances include vCPU and RAM in the hourly price; CPU-only groups are billed at $0.005 per vCPU-hour and $0.001 per GB-hour.
How reliable is an individual node?
Individual Community Cloud nodes run at roughly 90–95% reliability. Run at least three replicas, make your service tolerant of an instance disappearing without warning, and let SCE's reallocation do the rest; combined system reliability is far higher than any single node.
What should my container do differently?
Build for amd64. Support IPv6 if you use the Container Gateway. Don't ship an empty container without a long-running process. Expect longer cold starts than a data center and give instances a few minutes to pull large images. The docs have a full checklist.
Can I use my existing Kubernetes tooling?
Yes. The Virtual Kubelet integration lets you schedule pods onto SaladCloud as container groups without changing your pipelines.